- Content-based Recommendation -

Transcription

1 - Content-based Recommendation - Institute for Software Technology Inffeldgasse 16b/2 A-8010 Graz Austria 1

2 Content-based recommendation While CF methods do not require any information about the items, it might be reasonable to exploit such information; and recommend fantasy novels to people who liked fantasy novels in the past What do we need: some information about the available items such as the genre ("content") some sort of user profile describing what the user likes (the preferences) The task: learn user preferences locate/recommend items that are "similar" to the user preferences "show me more of the same what I've liked" 2

3 What is the "content"? Most CB-recommendation techniques were applied to recommending text documents. Like web pages or newsgroup messages for example. Content of items can also be represented as text documents. With textual descriptions of their basic characteristics. Structured: Each item is described by the same set of attributes attributes Title Genre Author Type Price Keywords The Night of the Gun Memoir David Carr Paperback Press and journalism, drug addiction, personal memoirs, New York The Lace Reader Fiction, Mystery Brunonia Barry Hardcover American contemporary fiction, detective, historical Into the Fire Romance, Suspense Suzanne Brockmann Hardcover American fiction, murder, neo-nazism Unstructured: free-text description. 3

4 Content-based recommendation Item representation Title Genre Author Type Price Keyw ords The Night of the Gun Memoir David Carr Paperback Press and journalism, drug addiction, personal memoirs, New York The Lace Reader Fiction, Mystery Brunonia Barry Hardcover American contemporary fiction, detective, historical Into the Fire Romance, Suspense Suzanne Brockmann Hardcover American fiction, murder, neo- Nazism User profile Title Genre Author Type Price Keywords Fiction Brunonia, Barry, Ken Follett Paperback Detective, murder, New York Simple approach Compute the similarity of an unseen item with the user profile based on the keyword overlap (e.g. using the Dice coefficient) Or use and combine multiple metrics 4

5 Example: Learning unit Recommendation Example set of software engineering related learning units (LU) Each of the learning units is additionally characterized by a set of categories (Java, UML, Management, Quality), for example, the learning unit l 1 is assigned to the categories Java and UML. 5

6 Example: Learning unit Recommendation Example of content-based filtering: user U a has already consumed the items l 4, l 6, and l 10. The item most similar to the preferences of U a is l 8 and is now the best recommendation candidate for the current user. 6

7 Term-Frequency - Inverse Document Frequency (TF IDF) Simple keyword representation has its problems in particular when automatically extracted not every word has similar importance longer documents have a higher chance to have an overlap with the user profile Standard measure: TF-IDF Encodes text documents in multi-dimensional Euclidian space weighted term vector TF: Measures, how often a term appears (density in a document) assuming that important terms appear more often normalization has to be done in order to take document length into account IDF: Aims to reduce the weight of terms that appear in all documents 7

8 TF IDF II Given a keyword i and a document j TF (i,j) term frequency of keyword i in document j IDF (i) inverse document frequency calculated as IDF i = log N / ni N : number of all recommendable documents n (i) : number of documents from N in which keyword i appears TF IDF is calculated as: TF-IDF i,j = TF i,j IDF i 8

9 Example TF-IDF representation Antony and Cleopatra Julius Caesar The Tempest Hamlet Othello Macbeth Antony Brutus Caesar Calpurnia Cleopatra mercy worser Vector v with dimension v = 7 9

10 Example TF-IDF representation Combined TF-IDF weights Each document is now represented by a real-valued vector of TF-IDF weights R v Antony and Cleopatra Julius Caesar The Tempest Hamlet Othello Macbeth Antony Brutus Antony and 0 Julius 1 The 0 Hamlet 0 Othello Macbeth Cleopatra Caesar Tempest Caesar Antony Calpurnia Brutus Cleopatra Caesar mercy Calpurnia worser Cleopatra mercy worser

11 Improving the vector space model Vectors are usually long and sparse remove stop words They will appear in nearly all documents. e.g. "a", "the", "on",... use stemming Aims to replace variants of words by their common stem e.g. "went" "go", "stemming" "stem",... size cut-offs only use top n most representative words to remove "noise" from data e.g. use top 100 words 11

12 Improving the vector space model II Use lexical knowledge, use more elaborate methods for feature selection Remove words that are not relevant in the domain Detection of phrases as terms More descriptive for a text than single words e.g. "United Nations" Limitations semantic meaning remains unknown example: usage of a word in a negative context "there is nothing on the menu that a vegetarian would like.. The word "vegetarian" will receive a higher weight then desired an unintended match with a user interested in vegetarian restaurants 12

13 Recommending Items Retrieval quality depends on individual capability to formulate queries with right keywords. Query-based retrieval: Rocchio's method The SMART System: Users are allowed to rate (relevant/irrelevant) retrieved documents (feedback) The system then learns a prototype of relevant/irrelevant documents Queries are then automatically extended with additional terms/weight of relevant documents 13

14 Rocchio details Document collections D + (liked) and D - (disliked) Calculate prototype vector for these categories. Computing modified query Q i+1 from currentquery Q i with: α, β, γ used to fine-tune the feedback α weight for original query β weight for positive feedback γ weight for negative feedback Often only positive feedback is used More valuable than negative feedback 14

15 Practical challenges of Rocchio s method Certain number of item ratings needed to build reasonable user model Can be automated by trying to capture user ratings implicitly (click on document) Pseudorelevance Feedback: Assume that the first n documents match the query best. The set D is not used until explicit negative feedback exists. User interaction required during retrieval phase Interactive query refinement opens new opportunities for gathering information and Helps user to learn which vocabulary should be used to receive the information he needs 15

16 Probabilistic methods Recommendation as classical text classification problem long history of using probabilistic methods Simple approach: 2 classes: hot/cold simple boolean document representation calculate probability that document is hot/cold based on Bayes theorem 16

17 Improvements Boolean representation simplistic positional independence assumed keyword counts lost More elaborate probabilistic methods e.g., estimate probability of term v occurring in a document of class C by relative frequency of v in all documents of the class Use other information retrieval methods (used by search engines..) 17

18 Explicit Decision Models Decision tree for recommendation problems inner nodes labeled with item features (keywords) used topartition the test examples existence or non existence of a keyword in basic setting only two classes appear at leaf nodes interesting or not interesting decision tree can automatically be constructed from training data works best with small number of features use meta features like author name, genre,... instead of TF-IDF representation. 18

19 Limitations of content-based recommendation Keywords alone may not be sufficient to judge quality/relevance of a document or web page up-to-date-ness, usability, aesthetics, writing style content may also be limited / too short content may not be automatically extractable (multimedia) Overspecialization Algorithms tend to propose "more of the same Or: too similar news items 19

20 Discussion & Summary In contrast to collaborative approaches, content-based techniques do not require user community in order to work Presented approaches aim to learn a model of user's interest preferences based on explicit or implicit feedback Deriving implicit feedback from user behavior can be problematic Danger exists that recommendation lists contain too many similar items All learning techniques require a certain amount of training data Some learning methods tend to overfit the training data Pure content-based systems are rarely found in commercial environments 20

21 Exercise Select an item domain that can be the basis for the implementation of a content-based filtering (CBF) recommender system. Develop an item table (including the corresponding item descriptions) T with at least 6 entries. On the basis of the CBF concepts discussed in the lecture unit exemplify the determination of contentbased filtering recommendations for an example user. 21

22 Literature [Michael Pazzani and Daniel Billsus 1997] Learning and revising user profiles: The identification of interesting web sites, Machine Learning 27 (1997), no. 3, [Soumen Chakrabarti 2002] Mining the web: Discovering knowledge from hyper-text data, Science & Technology Books, [A. Felfernig, M. Jeran, G. Ninaus, F. Reinfrank, and S. Reiterer], Basic Approaches in Recommendation Systems, Recommendation Systems in Software Engineering, Springer Verlag, pp Springer,

23 Thank You! 23